ERC-8004: Trustless Agents

Standardized on-chain discovery, reputation, and validation for AI Agents.

Overview

ERC-8004 introduces a comprehensive framework for "Trustless Agents" on EVM blockchains. It consists of three core registries:

1. Identity Registry: ERC-721 based system for Agent registration, discovery, and management.
2. Reputation Registry: On-chain feedback system for storing signals (uptime, success rate, user reviews).
3. Validation Registry: Cryptographic verification of agent work (zkML, TEE attestation signals).

Project Structure

• src/: Solidity smart contracts
• test/: Foundry unit tests
• script/: Deployment scripts
• Makefile: Command aliases

Prerequisites

• Foundry
• Git

Getting Started

1. Clone the repository:

   git clone https://github.com/EIPs-CodeLab/ERC-8004.git
   cd ERC-8004

2. Install dependencies:

   forge install

3. Build contracts:

   make build

4. Run tests:

   make test

Development

Running Tests

Run all tests:

forge test

Run specific test:

forge test --match-contract IdentityRegistryTest

Deployment (Sepolia)

1. Copy .env.example to .env (if not present) and fill in your keys:

   PRIVATE_KEY=your_private_key
   SEPOLIA_RPC_URL=https://rpc.sepolia.org
   ETHERSCAN_API_KEY=your_etherscan_key

2. Deploy:

   make deploy-sepolia

Real-World Use Cases

Feature	With ERC-8004	Without ERC-8004
Discovery	Universal registry (eip155:1:0x...) browseable by any generic dApp	Fragmented, proprietary databases or individual project repositories
Reputation	Permanent, verifiable history of an agent's performance (uptime, earnings)	Siloed rating systems (like Uber stars) locked in specific platforms
Monetization	Agents can own their identity (NFT) and transfer accumulated reputation	Selling an agent account violates TOS or is impossible (keys tied to email)
Trust	Cryptographically linked validation (TEEs, ZK proofs) providing "proof of work"	"Trust me bro" or relying on centralized API guarantees

Security

This project follows the EIP-8004 Security Considerations.

contracts are designed to be immutable and permissionless where possible. See SECURITY.md for more details.